Research map / Learning theory and algorithms / Kernels and regression theory
Rethinking Generalization of Neural Models: A Named Entity Recognition Case Study
Jinlan Fu, Pengfei Liu, Qi Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Correct human annotation errors identified using entity coverage ratio | NER models evaluated on CoNLL · ReCoNLL | F1: 93.03 -> 93.78 | Improved |
| Combine character-level FLAIR with word-level GloVe | LSTM-CRF NER · CoNLL | F1: 91.79 -> 93.03 | Improved |
| Use character-level pretrained embeddings rather than randomly initialized word embeddings | NER models · WNUT | F1: 14.97 -> 41.49 | Improved |
| Use FLAIR + GloVe rather than randomly initialized word embeddings | NER models · WNUT | F1: 14.97 -> 45.96 | Improved |
| Use ELMo or FLAIR rather than simpler embedding choices | NER models · CoNLL and WNUT, seen entities | F1 | No clear effect |
| Train with entities seen under the same label rather than entities with no same-label training occurrence | NER models · CoNLL and WNUT | F1: 98.56 -> 73.58 | Improved |
| Select source domains using entity and contextual coverage rather than adding training data indiscriminately | Cross-domain NER · WNUT | F1 | Improved |
| Increase the amount of training data without prioritizing source-domain relevance | Cross-domain NER · WNUT | F1 | Mixed |
| Use contextual coverage in addition to entity coverage when choosing training domains | Cross-domain NER · OntoNotes-wb | F1: 13.8 -> 21.8 | Improved |
| Use a different architecture or pretrained knowledge source | NER models · Hard cross-category cases | error ratio | No clear effect |
| Train on examples from categories with conflicting gradient directions | NER models · Overlapping entity categories | cross-category error ratio | Hurt |
Closest papers by meaning
- UniversalNER: Targeted Distillation from Large Language Models for Open Named Entity Recognition (ICLR 2024 · similarity 0.89)
- GuideNER: Annotation Guidelines Are Better than Examples for In-Context Named Entity Recognition (AAAI 2025 · similarity 0.86)
- A Reasoning Paradigm for Named Entity Recognition (AAAI 2026 · similarity 0.86)
- Three Heads Are Better than One: Improving Cross-Domain NER with Progressive Decomposed Network (AAAI 2024 · similarity 0.86)
- ConsistNER: Towards Instructive NER Demonstrations for LLMs with the Consistency of Ontology and Context (AAAI 2024 · similarity 0.86)
- PaDeLLM-NER: Parallel Decoding in Large Language Models for Named Entity Recognition (NeurIPS 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
